EP0405706A1 - Processor unit networks - Google Patents
Processor unit networks Download PDFInfo
- Publication number
- EP0405706A1 EP0405706A1 EP90301296A EP90301296A EP0405706A1 EP 0405706 A1 EP0405706 A1 EP 0405706A1 EP 90301296 A EP90301296 A EP 90301296A EP 90301296 A EP90301296 A EP 90301296A EP 0405706 A1 EP0405706 A1 EP 0405706A1
- Authority
- EP
- European Patent Office
- Prior art keywords
- processor
- processor unit
- units
- groups
- serial
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Withdrawn
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F15/00—Digital computers in general; Data processing equipment in general
- G06F15/16—Combinations of two or more digital computers each having at least an arithmetic unit, a program unit and a register, e.g. for a simultaneous processing of several programs
- G06F15/163—Interprocessor communication
- G06F15/173—Interprocessor communication using an interconnection network, e.g. matrix, shuffle, pyramid, star, snowflake
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F9/00—Arrangements for program control, e.g. control units
- G06F9/06—Arrangements for program control, e.g. control units using stored programs, i.e. using an internal store of processing equipment to receive or retain programs
- G06F9/46—Multiprogramming arrangements
Definitions
- the present invention relates to processor unit networks.
- processor units for example central processor units (C.P.U)
- CPU units for example central processor units (C.P.U)
- a loosely coupled multiprocessor is one where each CPU has exclusive access to its own memory. Communications between CPUs in a loosely coupled multiprocessor is only accomplished by the passing of messages, there is no sharing of memory between CPUs.
- processor unit network or loosely coupled multi-processor that can remain operation when there is a hardware fault.
- processor network has been formed using either parallel buses or "point-to-point" serial links.
- Parallel buses have a greater throughput in terms of the number of bytes of information transferred per second. However they require a large number of parallel interconnections which take up edge connectors on a printed circuit board. In order to provide fault tolerance, it is necessary to duplicate the bus which only serves to increase the requirement for edge connectors. Fault tolerance on duplicated buses is further complicated by having to ensure that no single fault can lead to the disabling of both buses. Certain faults on parallel buses can be difficult to diagnose. The number of CPUs which can be connected together on a parallel bus is generally limited to 20 or less.
- a processor unit network comprising a plurality of processor units, the processor units being arranged into a plurality of processor unit groups, each processor unit being arranged to be serially coupled to each other processor unit within its group, and parallel buses are provided, each bus being arranged to be connected to a respective processor unit from each group.
- the processor groups are arranged into a serial plane or a braided ring.
- the processor units may be central processor units with respective memory means arranged to form a loosely coupled multi-processor.
- processor units 1, 3, 5, 7, 9 are arranged into adjacent processor groups, each unit 1, 3, 5, 7, 9 are serially coupled by serial links to each other whilst parallel buses are provided to exclusively couple common processor units i.e. 3′ to 3 ⁇ to 3′′′.
- parallel buses are provided to exclusively couple common processor units i.e. 3′ to 3 ⁇ to 3′′′.
Abstract
A processor unit network suitable for forming a loosely coupled multi-processor using parallel bus connections and serial linkages. Processor units (1, 3, 5, 7, 9) being arranged into processor groups, each processor unit (1, 3, 5, 7, 9) of a processor group being serial linked to other members of the group and common processor units of each group being exclusively coupled to adjacent processor groups, for example processor unit 3˝ to processor units 3′ and 3‴.
Description
- The present invention relates to processor unit networks.
- It is known to connect processor units, for example central processor units (C.P.U), to form a loosely coupled multiprocessor. A loosely coupled multiprocessor is one where each CPU has exclusive access to its own memory. Communications between CPUs in a loosely coupled multiprocessor is only accomplished by the passing of messages, there is no sharing of memory between CPUs.
- There is a requirement to provide a processor unit network or loosely coupled multi-processor that can remain operation when there is a hardware fault. Typically, such a processor network has been formed using either parallel buses or "point-to-point" serial links.
- Parallel buses have a greater throughput in terms of the number of bytes of information transferred per second. However they require a large number of parallel interconnections which take up edge connectors on a printed circuit board. In order to provide fault tolerance, it is necessary to duplicate the bus which only serves to increase the requirement for edge connectors. Fault tolerance on duplicated buses is further complicated by having to ensure that no single fault can lead to the disabling of both buses. Certain faults on parallel buses can be difficult to diagnose. The number of CPUs which can be connected together on a parallel bus is generally limited to 20 or less.
- "Point-to-point" serial links have a lower data transfer rate compared parallel buses, but the overall throughput can be increased by having more than one link. However it is difficult to increase the number of links to large numbers because of edge connector limitations. Thus the number of CPUs which can be connected by a full mesh interconnect topology is limited to N + 1. (Where N is the number of links to each CPU). They do have the advantage that faults on the serial link can be isolated. If it is required to increase the number of CPUs above N + 1, then there are different topologies available for example Hypercubes and braided rings. The main disadvantage of these topologies is the need to relay messages via intermediate nodes. This is not such a problem when the number of intermediate nodes is relatively small. However as the network increases in size, it is possible to use up large proportions of both the bandwidth of each serial link and the processing power of intermediate nodes, with transit messages. Balancing the message traffic evenly over the many serial links is particularly difficult in such networks. This is exacerbated by the need to be able to tolerate the failure of links.
- It is an objective of the present invention to provide a processor unit network topology that substantially relieves the problems defined above.
- According to the present invention there is provided a processor unit network comprising a plurality of processor units, the processor units being arranged into a plurality of processor unit groups, each processor unit being arranged to be serially coupled to each other processor unit within its group, and parallel buses are provided, each bus being arranged to be connected to a respective processor unit from each group.
- Preferably, the processor groups are arranged into a serial plane or a braided ring. The processor units may be central processor units with respective memory means arranged to form a loosely coupled multi-processor.
- An embodiment of the present invention will now be described by way of example only with reference to the drawing, a schematic illustration of a processor unit network topography in accordance with the present invention.
- Referring to the drawing, a plurality of
processor units unit - Communication between CPUs in fault free conditions involves a maximum of one additional transit CPU. Routing between two CPUs in the general case would involve one serial link and one parallel bus. Choice of which to use first can be arbitrary, however if every CPU operates to the same rule, then the message load is automatically spread over the available serial links.
- Under fault conditions this topology offers the tradeoff between routing complexity and loss of processing capacity. Take the case of a failure of a parallel bus. This is unlikely to cause the failure of the serial links to the CPUs on the bus, so communications between these CPUs can be maintained by relaying messages via the remaining working buses. These messages can be distributed over a number of working buses so as to minimise the extra load. Alternatively, it is possible to adopt a strategy which treats failure of the bus as a failure of all the CPUs on the bus. However this does lead to the loss of a large proportion of the processing power of the system.
- In accordance with the present invention it is possible to establish a network comprising just two CPUs and additional CPUs added as and when required. Extension of the processor network can proceed in one of two dimensions depending on the requirements of the particular application. Starting with a pair of CPUs with a serial link between them (the minimum requirement for fault tolerance), it is possible to expand firstly by adding pairs of CPUs, one on each bus. Alternatively it is possible to expand the number of parallel buses first. The flexibility provided by having this ability is seen as being one of the advantages of this topology.
- It will be understood, that although the drawing illustrates the processor groups in a serial plane with a full mesh interconnect, that alternative topology could be used, in particular a braided ring.
Claims (4)
1. A processor unit network comprising a plurality of processor units, these processor units being arranged into a plurality of processor unit groups, each processor unit being arranged to be serially coupled to each other processor unit within its group, and parallel buses are provided, each bus being arranged to be connected to a respective processor unit from each group.
2. A processor unit network as claimed in claim 1 wherein the processor groups are arranged in serial planes.
3. A processor unit network as claimed in claim 1 wherein the processor groups are arranged in braided rings.
4. A processor unit network as claimed in any preceding claim wherein the processor units are central processor units with respective memory means arranged to form a loosely coupled multi-processor.
Applications Claiming Priority (2)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
GB8912465 | 1989-05-31 | ||
GB8912465A GB2232512A (en) | 1989-05-31 | 1989-05-31 | Processor unit networks |
Publications (1)
Publication Number | Publication Date |
---|---|
EP0405706A1 true EP0405706A1 (en) | 1991-01-02 |
Family
ID=10657627
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
EP90301296A Withdrawn EP0405706A1 (en) | 1989-05-31 | 1990-02-07 | Processor unit networks |
Country Status (9)
Country | Link |
---|---|
EP (1) | EP0405706A1 (en) |
JP (1) | JPH02308357A (en) |
KR (1) | KR900018810A (en) |
CN (1) | CN1047743A (en) |
AU (1) | AU5143190A (en) |
CA (1) | CA2011340A1 (en) |
FI (1) | FI902698A0 (en) |
GB (1) | GB2232512A (en) |
PT (1) | PT94200A (en) |
Cited By (9)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
WO2005053240A2 (en) * | 2003-11-19 | 2005-06-09 | Honeywell International Inc. | Relaying data in unsynchronous mode of braided ring networks |
US7372859B2 (en) | 2003-11-19 | 2008-05-13 | Honeywell International Inc. | Self-checking pair on a braided ring network |
US7656881B2 (en) | 2006-12-13 | 2010-02-02 | Honeywell International Inc. | Methods for expedited start-up and clique aggregation using self-checking node pairs on a ring network |
US7668084B2 (en) | 2006-09-29 | 2010-02-23 | Honeywell International Inc. | Systems and methods for fault-tolerant high integrity data propagation using a half-duplex braided ring network |
US7778159B2 (en) | 2007-09-27 | 2010-08-17 | Honeywell International Inc. | High-integrity self-test in a network having a braided-ring topology |
US7889683B2 (en) | 2006-11-03 | 2011-02-15 | Honeywell International Inc. | Non-destructive media access resolution for asynchronous traffic in a half-duplex braided-ring |
US7912094B2 (en) | 2006-12-13 | 2011-03-22 | Honeywell International Inc. | Self-checking pair-based master/follower clock synchronization |
US8817597B2 (en) | 2007-11-05 | 2014-08-26 | Honeywell International Inc. | Efficient triple modular redundancy on a braided ring |
CN104020756A (en) * | 2014-05-22 | 2014-09-03 | 国电南瑞科技股份有限公司 | Logic network topology sorting and storing method for fault diagnosing system |
Family Cites Families (10)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US3308436A (en) * | 1963-08-05 | 1967-03-07 | Westinghouse Electric Corp | Parallel computer system control |
NL8002787A (en) * | 1980-05-14 | 1981-12-16 | Philips Nv | MULTIPROCESSOR CALCULATOR SYSTEM FOR PERFORMING A RECURSIVE ALGORITHME. |
US4663706A (en) * | 1982-10-28 | 1987-05-05 | Tandem Computers Incorporated | Multiprocessor multisystem communications network |
US4814973A (en) * | 1983-05-31 | 1989-03-21 | Hillis W Daniel | Parallel processor |
GB2174519B (en) * | 1984-12-26 | 1988-09-01 | Vmei Lenin Nis | Multiprocessor system |
GB8528892D0 (en) * | 1985-11-23 | 1986-01-02 | Int Computers Ltd | Multi-node data processing system |
EP0232859A3 (en) * | 1986-01-27 | 1989-08-30 | International Business Machines Corporation | Processor intercommunication network |
NL8600218A (en) * | 1986-01-30 | 1987-08-17 | Philips Nv | NETWORK OF DATA PROCESSING STATIONS. |
GB8618943D0 (en) * | 1986-08-02 | 1986-09-10 | Int Computers Ltd | Data processing apparatus |
US4985832A (en) * | 1986-09-18 | 1991-01-15 | Digital Equipment Corporation | SIMD array processing system with routing networks having plurality of switching stages to transfer messages among processors |
-
1989
- 1989-05-31 GB GB8912465A patent/GB2232512A/en not_active Withdrawn
-
1990
- 1990-02-07 EP EP90301296A patent/EP0405706A1/en not_active Withdrawn
- 1990-02-26 KR KR1019900002458A patent/KR900018810A/en not_active Application Discontinuation
- 1990-03-02 CA CA002011340A patent/CA2011340A1/en not_active Abandoned
- 1990-03-19 AU AU51431/90A patent/AU5143190A/en not_active Abandoned
- 1990-04-10 CN CN90102150A patent/CN1047743A/en active Pending
- 1990-04-12 JP JP2097437A patent/JPH02308357A/en active Pending
- 1990-05-30 FI FI902698A patent/FI902698A0/en not_active Application Discontinuation
- 1990-05-30 PT PT94200A patent/PT94200A/en not_active Application Discontinuation
Non-Patent Citations (3)
Title |
---|
9TH ANNUAL SYMP. ON COMPUTER ARCHITECTURE, Austin, Texas, 26th - 29th April 1982, pages 90-98, IEEE, New York, US; L.N. BHUYAN et al.: "A general class of processor interconnection strategies" * |
IEEE INT. CONF. ON COMPUTER DESIGN: VLSI IN COMPUTERS, New York, 6th - 9th October 1986, pages 230-233, IEEE, New York, US; F. FRANZON: "Interconnect strategies for fault tolerant 2D VLSI arrays" * |
PATENT ABSTRACTS OF JAPAN, vol. 6, no. 245 (P-159)[1123], 3rd October 1982; & JP-A-57 143 657 (FUJITSU K.K.) 04-09-1982 * |
Cited By (15)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US7649835B2 (en) | 2003-11-19 | 2010-01-19 | Honeywell International Inc. | Unsynchronous mode brother's keeper bus guardian for a ring networks |
WO2005053240A2 (en) * | 2003-11-19 | 2005-06-09 | Honeywell International Inc. | Relaying data in unsynchronous mode of braided ring networks |
US7372859B2 (en) | 2003-11-19 | 2008-05-13 | Honeywell International Inc. | Self-checking pair on a braided ring network |
US7502334B2 (en) | 2003-11-19 | 2009-03-10 | Honeywell International Inc. | Directional integrity enforcement in a bi-directional braided ring network |
US7505470B2 (en) | 2003-11-19 | 2009-03-17 | Honeywell International Inc. | Clique aggregation in TDMA networks |
US7606179B2 (en) | 2003-11-19 | 2009-10-20 | Honeywell International, Inc. | High integrity data propagation in a braided ring |
WO2005053240A3 (en) * | 2003-11-19 | 2005-07-28 | Honeywell Int Inc | Relaying data in unsynchronous mode of braided ring networks |
US7729297B2 (en) | 2003-11-19 | 2010-06-01 | Honeywell International Inc. | Neighbor node bus guardian scheme for a ring or mesh network |
US7668084B2 (en) | 2006-09-29 | 2010-02-23 | Honeywell International Inc. | Systems and methods for fault-tolerant high integrity data propagation using a half-duplex braided ring network |
US7889683B2 (en) | 2006-11-03 | 2011-02-15 | Honeywell International Inc. | Non-destructive media access resolution for asynchronous traffic in a half-duplex braided-ring |
US7656881B2 (en) | 2006-12-13 | 2010-02-02 | Honeywell International Inc. | Methods for expedited start-up and clique aggregation using self-checking node pairs on a ring network |
US7912094B2 (en) | 2006-12-13 | 2011-03-22 | Honeywell International Inc. | Self-checking pair-based master/follower clock synchronization |
US7778159B2 (en) | 2007-09-27 | 2010-08-17 | Honeywell International Inc. | High-integrity self-test in a network having a braided-ring topology |
US8817597B2 (en) | 2007-11-05 | 2014-08-26 | Honeywell International Inc. | Efficient triple modular redundancy on a braided ring |
CN104020756A (en) * | 2014-05-22 | 2014-09-03 | 国电南瑞科技股份有限公司 | Logic network topology sorting and storing method for fault diagnosing system |
Also Published As
Publication number | Publication date |
---|---|
CN1047743A (en) | 1990-12-12 |
CA2011340A1 (en) | 1990-11-30 |
PT94200A (en) | 1992-02-28 |
KR900018810A (en) | 1990-12-22 |
AU5143190A (en) | 1990-12-06 |
GB8912465D0 (en) | 1989-07-19 |
GB2232512A (en) | 1990-12-12 |
JPH02308357A (en) | 1990-12-21 |
FI902698A0 (en) | 1990-05-30 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
US5321813A (en) | Reconfigurable, fault tolerant, multistage interconnect network and protocol | |
JP3301053B2 (en) | Interconnect network | |
US5751710A (en) | Technique for connecting cards of a distributed network switch | |
US20020040425A1 (en) | Multi-dimensional integrated circuit connection network using LDT | |
US20060282648A1 (en) | Network topology for a scalable multiprocessor system | |
EP0392216B1 (en) | Network rearrangement method and system | |
EP0405706A1 (en) | Processor unit networks | |
US7684328B2 (en) | Data transfer network | |
EP2095649B1 (en) | Redundant network shared switch | |
EP1367778A1 (en) | Networked computer system and method using dual bi-directional communication rings | |
Kumar et al. | Design and analysis of fault-tolerant multistage interconnection networks with low link complexity | |
CN114138354B (en) | Multi-host supporting on-board OCP network card system and server | |
US4048482A (en) | Arrangement for controlling a signal switching system and a method for using this arrangement | |
Agrawal et al. | A survey of communication processor systems | |
CN111782565A (en) | GPU server and data transmission method | |
CN113691433B (en) | Data transmission system, method, device, electronic equipment and storage medium | |
KR0150070B1 (en) | A hierarchical crossbar interconnection network for the cluster-based parallel processing computer | |
KR0170496B1 (en) | Cluster connecting structure using cross bar switch in a parallel processing computer system | |
KR0146562B1 (en) | A processor connecting apparatus applied the scheme of hierachical crossbar switch at a parallel computer system | |
KR0171015B1 (en) | Traffice dispersion control node apparatus | |
Biswas et al. | A centrally controlled shuffle network for reconfigurable and fault-tolerant architecture | |
EP0957611A2 (en) | Method of establishing a redundant mesh network using a minimum number of links | |
Raghavendra et al. | Reliability and fault-tolerance in multistage interconnection networks | |
Biswas et al. | Department of Electrical Communication Engineering, Indian Institute of Science, Bangalore 560 01 2, INDIA | |
Masuyama et al. | The Number of Permutations Realizable in Fault-Tolerant Multistage Interconnection Networks |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
PUAI | Public reference made under article 153(3) epc to a published international application that has entered the european phase |
Free format text: ORIGINAL CODE: 0009012 |
|
AK | Designated contracting states |
Kind code of ref document: A1 Designated state(s): BE DE DK ES FR GR IT LU NL SE |
|
17P | Request for examination filed |
Effective date: 19910418 |
|
STAA | Information on the status of an ep patent application or granted ep patent |
Free format text: STATUS: THE APPLICATION HAS BEEN WITHDRAWN |
|
RAP1 | Party data changed (applicant data changed or rights of an application transferred) |
Owner name: GPT LIMITED |
|
18W | Application withdrawn |
Withdrawal date: 19911003 |
|
R18W | Application withdrawn (corrected) |
Effective date: 19911003 |